NBER WORKING PAPER SERIES

FLEXIBLE EXCHANGE RATES AS SHOCK ABSORBERS

Sebastian EdwardsEduardo Levy Yeyati

Working Paper 9867http://www.nber.org/papers/w9867

NATIONAL BUREAU OF ECONOMIC RESEARCH1050 Massachusetts Avenue

Cambridge, MA 02138July 2003

We are grateful to Maria Lorena Garegnani, Mariano Lanfranconi and Luciana Moscoso for their excellentresearch assistance, and to participants at the 7th LACEA Annual Meeting for their helpful comments. Theviews expressed herein are those of the authors and not necessarily those of the National Bureau of EconomicResearch

©2003 by Sebastian Edwards and Eduardo Levy Yeyati. All rights reserved. Short sections of text not toexceed two paragraphs, may be quoted without explicit permission provided that full credit including ©notice, is given to the source.

Flexible Exchange Rates as Shock AbsorbersSebastian Edwards and Eduardo Levy YeyatiNBER Working Paper No. 9867July 2003JEL No. F30, F32

ABSTRACT

In this paper we analyze empirically the effect of terms of trade shocks on economic performance

under alternative exchange rate regimes. We are particularly interested in investigating whether

terms of trade disturbances have a smaller effect on growth in countries with a flexible exchange rate

regime, than in countries with a more rigid exchange rate arrangement. We also analyze whether

negative and positive terms of trade shocks have asymmetric effects on growth, and whether the

magnitude of these asymmetries depends on the exchange rate regime. We find evidence suggesting

that terms of trade shocks get amplified in countries that have more rigid exchange rate regimes. We

also find evidence of an asymmetric response to terms of trade shocks: the output response is larger

for negative than for positive shocks. Finally, we find evidence supporting the view that, after

controlling for other factors, countries with more flexible exchange rate regimes grow faster than

countries with fixed exchange rates.

Sebastian Edwards Eduardo Levy YeyatiUCLA Anderson Graduate School of Business Universidad Torcuato di Tella110 Westwood Plaza, Suite C508 ArgentinaBox 951481 ely@utdt.eduLos Angeles, CA 90095-1481and NBERsebastian.edwards@anderson.ucla.edu

1

I. Introduction

During the last few years economists’ views on exchange rate regimes have

evolved significantly. Fixed-but-adjustable regimes have lost adepts, while hard-pegs

and floating rates have gained in popularity. The discussion on the relative merits of

these two contrasting exchange rate systems has come to be known as the “two corners”

debate (Fischer 2001). Supporters of hard-pegs have argued that this type of regime

provides credibility and results in lower inflation, a more stable economic environment

and faster economic growth.1 Supporters of flexibility, on the other hand, have argued

that under floating exchange rates the economy has a greater ability to adjust to external

shocks.2 According to this view, which at least goes back to Meade (1951), countries

with a flexible exchange rate system will be able to buffer real shocks stemming from

abroad. This, in turn, will allow countries with floating rates to avoid costly and

protracted adjustment processes.3

In most models of open economies, real external shocks – including terms of trade

and real interest rate shocks – will result in changes in the equilibrium real exchange rate

(Obstfeld and Rogoff, 1995). If the nominal exchange rate is fixed, the adjustment in the

equilibrium real exchange rate will have to take place through changes in domestic

nominal prices and domestic wages. As Meade (1951, p. 201-02) argued early on, this

adjustment will be difficult in countries with a fixed exchange rate and inflexible money

wages. According to Meade (1951), in the presence of these rigidities the economy is

likely to benefit from what he called a “variable exchange rate” regime, or from what

we know today as a floating exchange rate system. He was careful to note, however, that

flexible exchange rates are not a panacea, and that there are indeed circumstances when

they may not help to accommodate external disturbances. This would be the case, for

1 Hard-peg regimes include currency boards, currency unions and dollarization. The growth effect is suppose to take place through two channels: (a) dollarization will mean lower interest rates, higher investment and, thus, faster growth. Dornbusch, for instance, (2001, p.240) has emphasized this channel, arguing that dollarization-induced lower interest rates are “conducive to investment and risk-taking, which translates into growth, and … a virtuous circle.” And (b), by eliminating exchange rate volatility, hard-pegs will encourage international trade and this, in turn, will result in faster growth. Rose (2000), and Rose and Van Wincoop (2001), among others, have emphasized this trade channel within the context of currency unions. On analytical aspects of dollarization see Calvo (1999) and Eichengreen and Haussman (1999). 2 In this paper we will use the terms “floating” and “flexible” exchange rate interchangeably. 3 Friedman (1953) was an early proponent of this view. The idea that hard pegs magnify external shocks acquired greater prominence in the aftermath of the Argentine currency and debt crisis of 2001-2002.

2

instance, if due to indexation or other mechanisms real wages are inflexible.4 This key

point has also been recognized by modern scholars that have analyzed the merits of

alternative exchange rate regimes (Dornbusch, 2001; Kenen, 2002).

Recently, a number of authors have argued that flexible exchange rate systems

will not be effective in countries where the private and public sectors have large foreign

currency-denominated liabilities (Eichengreen and Hausmann 1999). In this case, it has

been argued, it is even possible that a flexible exchange rate regime will amplify the

negative effects of terms of trade shocks. The reason for this is that in the presence of

“balance sheet” effects, the currency depreciation generated by the external shock will

generate (large) increases in the value of the debt expressed in domestic currency. This,

in turn, may trigger bankruptcies, lead the public sector to insolvency, and result in a

reduction in the rate of growth (Calvo 2000).

As the preceding discussion suggests, determining whether flexible exchange rate

regimes are indeed able to insulate the economy from external shocks, and contribute to

improving economic performance, is ultimately an empirical issue; it can only be

elucidated by analyzing the historical evidence.5 Surprisingly, there has been very little

empirical work on the relationship between exchange rate regimes and the way in which

terms of trade shocks affect growth and other measures of economic performance. In

fact, papers that have investigated empirically the way in which terms of trade

disturbances affect economic growth and growth volatility, have tended to ignore the role

of the exchange rate regime in the transmission process. A literature search using

EconLit indicates that 165 papers with the words “exchange rate regimes” and “growth”

in the title or abstract were published between 1969 and 2002. During the same period,

98 papers with the words “terms of trade” and “growth” were published. However, only

3 articles that had all three terms were published during this 33-year span.6

The purpose of this paper is to bridge this gap in the literature, and to analyze

empirically the effect of terms of trade shocks on economic performance under 4 In fact, Meade (1951, p. 203) explicitly said that “for the variable-exchange-rate mechanism to work effectively there must be sufficient divorce in movements in the cost of living and movements in money wage rates.” 5 Calvo (2000), for instance, has argued that if there are “dollarized liabilities” a flexible exchange rate regime may

3

alternative exchange rate regimes. We are particularly interested in investigating

whether, as supporters of exchange rate flexibility have claimed, terms of trade

disturbances have a smaller effect on growth in countries with a flexible exchange rate

regime, than in countries with a more rigid exchange rate arrangement. We also analyze

whether negative and positive terms of trade shocks have asymmetric effects on growth,

and whether the magnitude of these asymmetries depends on the exchange rate regime.

In order to investigate these issues we use a new data set that provides an improved

classification of the exchange rate regime in each country at any particular moment in

time. The advantage of this data set – which was constructed by Levy Yeyati and

Sturzenegger (2002) – is that it does not rely on official country statements for classifying

countries as having a pegged, intermediate or floating regime.7 Instead, this new data set

uses actual data on the behavior of nominal exchange rates and international reserves to

classify countries under different regimes.

Our findings may be summarized as follows: First, we find evidence suggesting

that terms of trade shocks get amplified in countries that have more rigid exchange rate

regimes. Another way of saying this is that we find evidence indicating that, with other

things given, countries with flexible exchange rates are able to accommodate better real

external shocks. Second, we find evidence of an asymmetric response to terms of trade

shocks. More precisely, the output response is larger for negative than for positive

shocks, a fact consistent with the presence of asymmetries in price responses (with

downward nominal inflexibility leading to larger quantity adjustments). Interestingly,

while the output response in both directions is, again, larger the more rigid the exchange

rate, this asymmetry is not present under flexible regimes.8 In addition, we find evidence

supporting the view that, after controlling for other factors, countries with more flexible

exchange rate regimes grow faster than countries with fixed exchange rates, confirming

previous findings in Levy Yeyati and Sturzenegger (2003). 6 Broda (2001) is a recent contribution that analyzes whether the exchange rate regime makes a difference in the way in which terms of trade shocks impact economic performance. 7 It is well known that in many countries the authorities systematically state that they have a particular regime, when in reality they have a different one. See Edwards (1993) for a discussion on this issue. 8 This would be in principle consistent with the presence of fear of floating, as reflected in a partial response of nominal exchange rates to positive shocks that result in larger real contractions. This hypothesis or, more generally, the hypothesis that exchange rates elasticity tends to be smaller in the event of negative shocks, is a fruitful topic for future research.

4

The rest of the paper is organized as follows: In section II we present our

empirical framework, we discuss our data, and we present our basic results. In Section

III we examine the robustness of the results to the use of the IMF de jure classification,

and we extend the analysis to explore potential asymmetries in the output response to

terms of trade shocks. Finally, in Section IV we present some concluding remarks.

II. Terms of Trade Shocks, Exchange Rate Regimes and Growth: An Empirical

Analysis

Economists’ concerns with the effects of terms of trade changes on economic

growth go back, at least, to the writings of Prebisch (1950) and Singer (1950). These

influential authors made two claims: first, they argued that developing countries’ terms of

trade had exhibited a secular deterioration through time. And second, they argued that

this decline in relative exports’ prices contributed to the developing countries’ lack of

industrialization, and resulted in low rates of growth and further impoverishment. As a

result of Prebisch and Singer’s empirical propositions, a number of authors developed

theoretical models on the connection between terms of trade and economic growth. The

majority of these models considered rather simple links, and argued that by negatively

affecting real income, negative terms of trade shocks depressed aggregate demand and,

thus, resulted in lower growth (Bloomfield 1984, Singer and Lutz 1994). More recent

studies, however, have focused on a variety of transmission channels, including the effect

of terms of trade on relative prices. Barro and Sala-I-Martin (1995), for example, have

pointed out that whether growth in fact accelerates as a result of terms of trade

improvements depends on the effects of relative price changes on productivity

improvements.

Some authors have emphasized the effects of terms of trade shocks on capital

accumulation and factor intensities. Basu and McLeod (1992), for example, constructed a

stochastic model of growth where imported intermediate inputs are complementary to

capital. In this setting, deterioration in the terms of trade makes imported inputs more

expensive and has the potential of reducing capital’s productivity. In addition, in this

model, uncertainty regarding the terms of trade has a negative effect on investment and,

5

ultimately, on growth. Mendoza (1997) developed a stochastic model of growth in which

terms-of-trade uncertainty affects savings and growth. In this model terms of trade

improvements have a positive effect on savings, capital accumulation and, thus, on the

average rate of growth. The model also predicts that higher terms-of-trade variability

could result in either faster or slower growth, depending on the degree of risk aversion.

In a comprehensive study Hadass and Williamson (2001) have reviewed most of

the empirical literature on terms of trade and economic performance produced during the

last five decades, including the works by Easterly et al (1993), Collier and Gunning

(1999), Warner (1992), Barro and Sala-I-Martin (1995) and Barro (1997). They

convincingly argue that, while there has been massive amount of work trying to explain

the actual behavior of terms of trade, relatively few studies have focused on the effects of

terms of trade shocks on growth. And none of the studies reviewed by them makes a

distinction between countries with different exchange rate regimes. In a recent

contribution, Broda (2001) provides one of the few empirical analyses on how terms of

trade shocks affect real economic performance under alternative exchange rate regimes.

He uses a VARs analysis to compute the way in which terms of trade shocks affect

growth. He finds that the (negative) effect of a 10% deterioration in the terms of trade

has a greater negative effect on growth under fixed than under flexible exchange rate

regimes.

II.1 The Empirical Model

Our main interest is to investigate whether, as supporters of floating exchange

rates have claimed, countries with floating exchange rate regimes are (partially) insulated

from the effects of terms of trade shocks on growth. More specifically, we are interested

in finding out if terms of trade disturbances affect differently countries with different

exchange rate regimes. The point of departure of our empirical analysis is a two-equation

formulation for the dynamics of real GDP per capita growth of country j in period t.

Equation (1) is the long run GDP growth equation, while equation (2) captures the

growth dynamics process.

6

(1) g* j = α + x j β + r j θ + ω j.

(2) ∆ g t j = λ [ g* j – g t-1 j ] + ϕ v t j + γ u t j + ξ t j .

The following notation has been used: g* j is the long run rate of real per capita GDP

growth in country j. x j is a vector of structural, institutional and policy variables that

determine long run growth; r j is a vector of regional dummies. α, β and θ are

parameters, and ω j is an error term assumed to be heteroskedastic. In equation (2), g t j is

the rate of growth of per capita GDP in country j in period t. The terms v t j and u t j are

shocks, assumed to have zero mean, finite variance and to be uncorrelated among

themselves. More specifically, v t j is assumed to be an external terms of trade shock,

while u t j captures other shocks, including political shocks. ξ t j is an error term, which is

assumed to be heteroskedastic – see equation (3) below for details. λ, ϕ, and γ are

parameters that determine the particular characteristics of the growth process.

From the perspective of the exchange rate regime discussion, an important

question is whether the exchange rate system has a direct effect on the long-term rate of

growth. We deal with this issue by investigating whether in equation (1) the intercept α is

different for countries with different exchange rate regimes.9 Equation (2) -- which has

the form of an equilibrium correction model (ECM) --, states that the actual rate of

growth in period t will deviate from the long run rate of growth due to the existence of

three types of shocks: v t j, u t j and ξ t j. Over time, however, the actual rate of growth will

tend to converge towards it long run value, with the rate of convergence given by λ.

Parameter ϕ, in equation (2), is expected to be positive, indicating that an improvement in

the terms of trade will result in a (temporary) acceleration in the rate of growth, and that

negative terms of trade shock are expected to have a negative effect on g t j.

Our main interest is to determine whether parameter ϕ in equation (3) depends on

the exchange rate regime of the country in question. If, as their supporters have argued,

floating exchange rates allow countries to absorb foreign shocks better, we would expect

ϕ to be smaller in countries with floating rates than in those countries with some version

7

of a pegged rates exchange rate regime. We are also interested in determining whether

positive and negative terms of trade shocks have asymmetric effects on growth – that we

do in section III. Our task, then, is to estimate the system given by equations (1) and (2),

and to analyze if the coefficients α and ϕ are different across exchange rate regimes. The

estimation of this system is not trivial, and is subject to the complexities of estimating

panels with lagged dependent variables and heteroskedastic errors.

We estimate the system (1) - (2) using a two-step procedure. In the first step we

estimate the long run growth equation (2) using a cross-country data set. These data are

averages for 1974-2000, and the estimation makes a correction for heteroskedasticity.

These first stage estimates are then used to generate long run predicted rates of growth to

replace g*j in the equilibrium correction model (2). In the second step, we estimate

equation (2) using a feasible generalized least squares procedure (FGLS) suggested by

Beck and Katz (1995) for unbalanced panels. In the estimation of equation (2) the error ξ t

j is assumed to be heteroskedastic, with a different variance for each of the k panels.

σ2

1 I 0 … 0 0 σ2

2 I … 0 (3) E [ξξ ’ ] = . . . . . . . . 0 0 … σ2

k I

The FGLS estimator has the same properties as the GLS estimator, and is

asymptotically efficient. Notice that an alternative estimation strategy would be to re-

parameterize equation (1) and (2), and to apply the Generalized-Method-of-Moments

(GMM) for dynamic panel data models suggested by Arellano and Bond (1991). When

we did this, the results obtained were similar to those obtained using our two-steps-FLS

methodology.10

9 On debates on the effect of alternative exchange rate regimes on performance see, for example, Gosh et al (1995), Levy Yeyati and Sturzenegger (2002), Frankel (1999) and Kenen (2002). 10 A potential limitation of the GMM strategy, however, is it does not lend itself to a straight forward interpretation of the equilibrium correction term. The results we obtained when using this method are available on request.

8

We use two alternative methods to investigate whether the terms of trade

coefficient ϕ in equation (2) is different for different exchange rate regimes: The first one

consists of including a variable that interacts the terms of trade shock with three

alternative indicators for exchange rate regimes. Our second method consists of splitting

the sample according to the exchange rate regime, and comparing the estimated

coefficients for the terms of trade variable. If flexible regimes buffer the country better

form external disturbances, we would expect the coefficient for the terms of trade

variable to be significantly lower in countries with flexible exchange rate regimes than in

countries with rigid regimes.

II.2 Data and Equation Specification

Our sample covers annual observations for 183 countries over the period 1974-

2000. With the exception of the civil unrest, exchange rate regimes, and secondary school

enrollment variables, the data were obtained from the IMF and the World Bank

databases. Since data availability varies across countries and periods, all tests were run

on consistent subsamples of observations corresponding to 96 and 100 countries. A list of

countries, as well as the definitions and sources for the variables used, are reported in

Appendix A.

As pointed out above, our main interest is to analyze the transmission of terms of

trade shocks under alternative exchange rate regimes. There is generalized agreement,

however, that the IMF’s “official” exchange rate regime classification tends to be

misleading. For this reason, we use a methodology proposed by Levy Yeyati and

Sturzenegger (2002) to construct four indexes of exchange rate regimes. These indexes

are constructed as time series, and are based on actual, as opposed to legal, exchange rate

behavior– see Appendix B for details. The three indexes are defined as follows:

• A binary index that takes the value of one if in that particular year the country

has a pegged exchange rate regime, and zero otherwise. We call this index

pegged.

• A dummy variable that takes the value of one if in that particular year the

country in question has a hard (as opposed to a conventional) peg – that is, if

9

it has a currency board, belongs to a currency union or it is dollarized. The

index takes a value of zero otherwise. This variable is called hard.

• A dummy variable that takes the value of one if the exchange rate regime is an

intermediate regime – crawling pegs, managed floats, and the like (see Levy

Yeyati and Sturzenegger 2002, for details). We call this index intermediate.

(Notice that from these definitions we are able to construct an index that takes

the value of one if the regime is neither pegged nor intermediate. This index

is called flexible.)

• And a three-way classification that combines some of the indexes described

above, and distinguishes between pegged, intermediate and flexible exchange

rate regimes. This index is called regime and takes a value of zero if in that

particular year the country in question has a flexible rate. It takes a value of

one if the country has an intermediate regime, and a value of two if the

country has a pegged exchange rate in that particular year.

These indexes where constructed for each year in the sample (1974-2000). Table 1

presents a summary of the distribution of countries in our sample across the different

exchange rate regimes that we have defined.

[TABLE 1 APPROXIMATELY HERE]

In estimating equation (1) for long run per capita growth, we follow the by now

standard literature on growth, as summarized by Barro and Sala-I-Martin (1995), and use

average data for 1974-2000. In terms of the equation specification, we follow Barro and

Sala-I-Martin (1995), Sachs and Warner (1995) and Dollar (1992) among others, and

assume that the rate of growth of GDP (g*j ) depends on a number of structural, policy

and social variables. More specifically, we include the following covariates: the log of

initial GDP per capita (gdpin); the investment ratio (invgdp); the coverage of secondary

education (sec, a proxy for human capital); an index of the degree of openness of the

economy (openness); the ratio of government consumption relative to GDP (gov); and

regional dummies for Latin American, Sub Saharan African and Transition economies

10

(latam, safrica, and trans). In some specifications we also included the rate of growth of

population. Finally, and in order to investigate whether the exchange rate regime affects

long run growth, in some of the cross-section regressions we also incorporated two

alternative indexes for the exchange rate regime. The first one, which we call

pegged_cross, is a cross section version of the time series index pegged, defined above.

It takes the value of one if the country in question has been classified as a fixed exchange

rate regime for at least 50% of the time, and zero otherwise. The second index – called

regime_cross – is a cross section version of the index regime described above, and is

constructed as an average of that index. A lower value of this regime_cross index, then,

represents a more flexible exchange rate regime. A full description of the data, including

the data sources, is provided in Appendix A.

In the estimation of dynamics of growth equation (2), v t j is the terms of trade

shock (∆tt), and is defined as the percentage change of the relative price of exports to

imports. Thus, a positive (negative) number represents an improvement (deterioration) in

the terms of trade. In addition to the terms of trade shocks, we also included the terms of

trade shock interacted with our different exchange rate regime indexes. An index of civil

unrest was included as a proxy for other shocks (this can be interpreted as being an

element in vector u t j in equation (1)). In all equations we included time fixed effects,

which capture systemic shocks to all countries, such as changes in global liquidity and

shocks to world interest rates. We also included regional dummies, and in some of the

equations we included lagged values of the terms of trade shocks. In Table 2 we present

summary statistics for all variables used in our empirical analysis (See Appendix A for

data sources).11

[TABLE 2 APPROXIMATELY HERE]

II.3 Main Results

The results from the first step estimation for the long run growth equation are

reported in Table 3, where the t-statistics have been estimated using robust standard

11 As usual, data availability differs across countries and variables. For consistency, the statistics reported in the table are based on the actual sample used in the empirical tests below.

11

errors computed using the Huber-White methodology. As may be seen, the results are

quite satisfactory; all the coefficients have the expected sign and most of them are

statistically significant. These results confirm previous findings with respect to the roles

played by initial GDP, education, openness, and government consumption in explaining

differentials in long run GDP per capita growth across countries. In terms of the main

question raised in this paper, a particularly interesting finding in Table 3 is that the

estimated coefficients for our two exchange rate regime indicators are significantly

negative. This suggests that in the long run, and after controlling for traditional

covariates, countries with (de facto) more rigid exchange rate regimes have tended to

grow at a slower rate than countries with more flexible exchange rate systems. Moreover

the absolute values of the point estimates are quite large, suggesting that, with other

things given, countries with a fixed exchange rate regime have had a lower rate of growth

of GDP per capita ranging between 0.66 and 0.85 percentage point per year, than

countries with a flexible regime. These findings contrast with studies such as Gosh et al

(1995) and IMF (1997) that have used the official IMF classification of regimes.

According to these studies, while fixed exchange rate countries tend to have a lower rate

of inflation than flexible rate ones, there is no statistical difference in terms of GDP per

capita growth across both groups of countries. Our results, on the other hand, are

consistent with recent findings by Levy Yeyati and Stuzenegger (2003).

[TABLE 3 APPROXIMATELY HERE]

We use the fitted values from the estimates for long run GDP per capita growth

reported in equation (i) in Table 2 to construct a proxy for g*j in the second step

estimation of equation (2). When alternative specifications for the long run growth

equation were used, the results were very similar to those reported in the paper.12

Table 4.a contains the results from the estimation, using the FGLS procedure

described above, of several versions of equation (2) on the dynamics of growth. All

equations were estimated for the 1974 – 2000 period, and included yearly fixed effects;

this allows us to capture the effects of shocks that are systemic to all countries in a

12 They are available from the authors on request.

12

particular year, such as changes in international interest rates. As may be seen the results

are quite satisfactory. The estimated coefficient of [ g* j – g t-1 j ] is, as expected, positive,

significant, and smaller than one. The point estimates are on the high side -- between

0.75 and 0.79. For instance, according to equation (i) in Table 4.a, after 4 years

approximately 90% of a unitary shock to real GDP growth will be eliminated. Also, as

expected, the estimated coefficients of the terms of trade shock are always positive and

statistically significant, indicating that an improvement (deterioration) in the terms of

trade results in an acceleration (de-acceleration) in the rate of growth of real per capita

GDP. The results in Table 4.a also show that the coefficients of our political shocks

variable – civil unrest – are negative in every specification. However, they are not

significant at conventional levels.

Our main interest in this paper is the estimated coefficient of the interactive terms

between our exchange rate regime indexes and the terms of trade shock. As may be seen

from Table 4.a, the estimated coefficients of these interactive terms are always positive,

and in most regressions they are significant at conventional levels. This indicates that the

effects of terms of trade shocks on growth are larger under fixed exchange rate regimes

that under floating regimes. Consider, as an example, the case of equation (i) in Table

4.a: the terms of trade coefficient has a point estimate of 0.043; the estimated interactive

term, on the other hand, has a point estimate of 0.037. These results suggest, then, that in

pegged exchange rates countries a 10% deterioration in the international terms of trade

has been associated, on average, with a (contemporaneous) decline in GDP per capita

growth of 0.80 of one percentage point. In flexible exchange rates countries, on the other

hand, the same 10% decline in the international terms of trade has been associated on

average with a (contemporaneous) reduction in GDP per capita growth of 0.43 of one

percentage point. That is, according to this equation under flexible exchange rates the

effects of terms of trade shocks on growth are approximately one half than under pegged

regimes.

[TABLES 4.a THROUGH 4.c APPROXIMATELY HERE]

13

Table 4.b contains separate FGLS regression results for four groups of countries,

each corresponding to a different exchange rate regime. The first group is comprised of

countries that according to our indicator have a flexible exchange rate; the second sub-

sample contains countries with an intermediate regime. The third sub-sample contains

countries with a pegged exchange rate regime. And finally, the fourth sub-sample

corresponds to countries that according to our classification have had a hard peg regime.

As may be seen, the results indicate that the estimated coefficient of the terms of trade

variables is always positive and significant. What is particularly interesting from the

point of view of this paper’s topic is that point estimates of these coefficients are different

across the four sub-samples, and that they increase with the rigidity of the exchange rate

regime. Indeed, the sum of the contemporaneous and lagged terms of trade coefficients is

highest for the hard-peg regimes (0.168); the second highest value corresponds to the

pegged regimes (0.129). The sum of these coefficients is 0.071 for the intermediate

systems, and it is the lowest (0.057) for the group of countries that has had a flexible

exchange rate system. Moreover, as the χ2 tests reported in Table 4.c indicate, the

coefficients for pegs are significantly larger (from a statistical point of view) than those

for each of the more flexible regimes.13

In order to investigate further how of terms of trade shocks affect growth under

alternative regimes, we divided our sample into industrial and emerging countries. This

allows us to analyze whether the results reported above are driven by the level of

development, rather than by their exchange rate regime. The results obtained – which are

reported in Table 5 –, show that flexible exchange rate regimes have helped buffer terms

of trade shocks for both industrial and emerging nations.

[TABLE 5 APPROXIMATELY HERE]

The results reported in Tables 4 and 5, then, provide support the hypothesis that

countries with flexible regimes have been able to accommodate terms of trade shocks 13 The χ2 statistics in this table were computed interacting each of the regressors with the corresponding regime dummy. Thus, for example, to compute the statistics for the pegged – flexible comparison we restricted the sample to include pegs and flexible regimes, interacted all controls with the pegged and

14

better than countries with rigid exchange rates. In the next section we expand our

analysis by investigating whether terms of trade shocks affect growth asymmetrically.

More precisely, we examine whether the impact of negative shocks is stronger than that

of positive shocks, as one should expect if nominal prices are rigid downward.

III. Asymmetric Effects and Robustness Analysis

In this section we deal with two extensions: First, we investigate whether positive

and negative terms of trade shocks affect growth in an asymmetric way, and whether

these asymmetric effects are different under alternative exchange rate regimes. And

second, we analyze the robustness of our results to alternative classifications of exchange

rate regimes, and to the use of alternative samples.

III.1 Asymmetric Effects of Terms of Trade Shocks under Alternative Regimes

According to a number of authors the most important advantage of flexibility is

that it allows the economy to buffer negative terms of trade shocks through smooth

changes (depreciations) in the real exchange rate. This contrasts with the case of pegged

exchange rates, where real exchange rate depreciation requires a decline in nominal

prices. If nominal prices are rigid downward, however, a negative terms of trade shock

will result in unemployment, a decline in output and I the rate of growth – see Dornbusch

(2001) and Kenen (2002) for details.In that sense, then, iIt is possible to argue that from a

policy point of view what really matters is the way in which alternative exchange rate

regimes accommodate negative terms of trade shocks.

In order to investigate this issue, we estimated a number of regressions that

distinguished between positive and negative terms of trade shocks. As before we used a

FGLS procedure for heteroskedastic panels. In this case, our system becomes:

(3) g* j = α + x j β + r j θ + ω j.

(4) ∆ g t j = λ [ g* j – g t-1 j ] + ϕ vp t j + ψ vn t j + γ u t j + ξ t j .

flexible dummies, and tested the null ∆tt* pegged+∆tt_1*neg*pegged – (∆tt*flexible + ∆tt_1*flexible) = 0. The last column (pegged vs. hard) compares conventional and hard pegs regimes.

15

Where vp t j refers to positive terms of trade shocks, vn t j refers to negative shocks; ϕ and

ψ are coefficients to be estimated. If the effects of negative terms of trade disturbances

on growth are indeed larger than those of positive shocks, we would expect ψ to be

significantly larger than ϕ. In the estimation of equations (3) and (4) we made a

distinction between four exchange rate regimes: hard-pegged; pegged; intermediate and

flexible. If, as its supporters have argued, flexible regimes are able to accommodate better

negative real shocks from abroad, the estimated ψs – that is, the coefficients of the

negative terms of trade shocks -- would be larger in countries with more rigid exchange

rate regimes than in countries with more flexible ones. As before, in the estimation of the

equation on growth dynamics (equation (4)) we included time specific effects.

The results from the estimation of the second stage equation (4) are presented in

Table 6 for two alternative samples: one that includes all countries and a sub-sample of

emerging countries only. As may be seen, the results obtained indicate that there are

indeed asymmetric effects of terms of trade shocks in six out of the seven regressions. In

every equation, with the exception of (i), the sum of the coefficients for the negative

shocks is higher than the sum of the coefficients for the positive shocks. Moreover, the

differences in the terms of trade coefficients are statistically significant for the hard pegs,

pegs, and intermediate regimes – see table 7 for formal tests.

[TABLE 6 APPROXIMATELY HERE]

[TABLE 7 APPROXIMATELY HERE]

Consider, for example, the comparison between pegged and flexible regimes for

the whole sample. According to the FGLS estimates in Table 6, the sum of the

coefficients of the negative terms of trade shocks is 0.158 for the pegged regime

countries, but only 0.053 for those countries with a flexible exchange rate. From a

statistical point of view, the sum of the (negative) terms of trade coefficients is

significantly higher for the pegged regime countries – the χ2 has a p-value of 0.017

(Table 7).14 In fact, similar tests indicate that the sum of the negative terms of trade

14 The χ2 statistics in this table were computed interacting each of the regressors with the corresponding regime dummy. Thus, to compute the statistics for negative shocks for the pegged – flex comparison we

16

coefficients for countries with flexible (fixed) exchange rate arrangements are

significantly lower (higher) than the sum of the coefficients for countries with non-

flexible (non-pegged) rates. Thus, these results indicate that the reported asymmetric

response to shocks in countries with more rigid exchange rate regimes is mainly driven

by the larger effects of negative terms of trade shocks. This, combined with the fact that

the sensitivity to both positive and negative shocks is higher the less flexible the regime,

suggests that the lack of exchange rate flexibility increases the real impact of terms of

trade shocks due to the lack of (downward) price flexibility.

To summarize, these results provide further support for the hypothesis that

flexible exchange rates have played a role as shock absorbers, helping countries

accommodate real terms of trade shocks. This ability to accommodate these shocks

appears to have been particularly important in the presence of negative external shocks.

III.2 Alternative Classification of Exchange Rate Regimes

In this subsection we investigate whether the results reported above depend on the

classification of exchange rate regimes that we have used. In order to do this we re-

estimated our model using the standard and official exchange rate classification provided

by the IMF. The results obtained in this case, not reported here due to space

considerations, are somewhat weaker.15 Although the coefficients have the expected

signs, and in most cases have similar point estimates to those reported in Tables 4-6, they

are estimated with a lower degree of precision.

To explore further this issue we conducted the following simple exercise: we

revised the IMF-based classification, and tried to detect obvious misclassifications of

regimes. We then re-estimated our equations using a restricted sample that include only

uncontroversial de jure IMF-defined regimes. This entails the (relatively minor) loss of

89 observations.16 The estimates, which are available from the authors on request, have a

higher degree of precision than those obtained when the unadjusted IMF classification is

used. As before, these results indicate that terms of trade shocks – and in particular included pegs and flexible regimes, interacted all the controls with the pegged and flexible dummies, and tested the null ∆tt*neg*pegged+∆tt*neg_1*pegged – (∆tt*neg*flexible + ∆tt*neg_1*flexible) = 0. 15 These results are available on request.

17

negative terms of trade shocks – have a larger effect on growth under rigid exchange rate

regimes than under more flexible regimes.

IV. Concluding Remarks

In this paper we examined two aspects of the economic implications of exchange

rate regimes that, despite being recurrently used to argue in favor of exchange rate

flexibility, have been the subject of little, if any, empirical work: (i) the role played by

flexible exchange rates as absorbers of real shocks, and (ii) the link between this role and

the presence of downward price rigidities. More precisely, we tested whether the

sensitivity of real growth to terms of trade shocks declines as the degree of flexibility of

the regime increases. In addition, we investigated whether this sensitivity is higher in the

event of negative shocks, as it would be the case in the presence of asymmetric price

rigidities.

Using a de facto classification of exchange rate regimes, we found that flexible

exchange rate arrangements indeed help reduce the real impact of terms of trade shocks

on GDP growth, both in emerging and industrial economies.17 Moreover, we found real

output growth to be more sensitive to negative than to positive shocks. In fact, most of

the differential shock responses across regimes can be traced to the stronger real impact

of negative shocks under a peg, be it of the conventional or the hard kind. The effects

unveiled in this paper are, on the other hand, not only statistically significant but

economically important: while a 10 percent deterioration of the terms of trade translates

into a real contraction of around 0.4% for the average country, this effect nearly doubles

under a peg. Thus, the choice of exchange rate regime indeed has important implications

in terms of output volatility. Moreover, the fact that the asymmetry of output responses

to real shocks increases with the rigidity of the regime suggests that pegs are associated

with deeper and longer contractions.

16 While a similar correction can be done for floats and intermediates, it is certainly in the fix group where misclassifications are less debatable, as changes in the exchange rate are readily observable. 17 Similar, albeit slightly weaker, results are obtained if a de jure regime classification is used.

18

References: Arellano, M., Bond, S., 1991, Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations, Review of Economic Studies, 58, 277-297. Barro, R. J., 1997, Determinants of Economic Growth, Cambridge, Massachussets, MIT Press. Barro, R., Sala-I-Martin, X.,1995, Economic Growth, McGraw Hill. Basu, P., McLeod, D., 1992, Terms of Trade, and Economic Fluctuations in Developing Countries, Journal of Development Economics, Vol. 37, pp. 89–110. Beck, N., Katz, J., 1995, What to Do (and not to Do) with Time-Series Cross-Section Data, American Political Science Review. Bloomfield, A., 1984, Effect of Growth on the Terms of Trade: Some Earlier Views, Essays in the history of international trade theory, Aldershot, UK. Broda, C., 2001, Coping with Terms of Trade Shocks: Pegs vs Floats, American Economic Review, 91(2), pp. 376-380. Calvo, G. A., 1999, Fixed vs. Flexible Exchange Rates. Preliminaries of a Turn-of-Millennium Rematch, University of Maryland. Mimeo. Calvo, G., 2000, The Case for Hard Pegs in the Brave New World of Global Finance, Mimeo, University of Maryland. Collier, P., Gunning, J.W., 1999, Why Has Africa Grown Slowly?, Journal of Economic Perspectives 13:3-22. Dollar, D., 1992, Exploiting the Advantages of Backwardness: The Importance of Education and Outward Orientation, Mimeo, World Bank. Dornbusch, R., 1980, Stabilization Policy in the Open Economy, Joint Economic Committee, Special Study on Economic Change, Washington D.C.. Dornbusch, R., 2001, Fewer Monies Better Monies, Discussion on Exchange Rates and the Choice of Monetary-Policy Regimes, The American Economic Review, ppg 238-242. Easterly, W., Kremer, M., Pritchett, L., Summers, L.H., 1993, Good Policy or Good Luck? Country Growth Performance and Temporary Shocks, Journal of Monetary Economics, 32:459-83.

19

Edwards, S., 1993, Openness, Trade Liberalization, and Growth in Developing Countries, Journal of Economic Literature, 1358-1393. Eichengreen, B., Haussman, R., 1999, Exchange Rates and Financial Fragility, NBER Working Paper No. 7418. Fischer, S., 2001, Exchange Rate Regimes: Is the Bipolar View Correct?, Distinguished Lecture on Economics in Government, AEA Meetings, New Orleans. Frankel, J., 1999, No single Currency Regime is Right for all Countries or at all Times, NBER Working Paper No. 7338. Friedman, M., 1953, The Case For Flexible Exchange Rates, in Essays in Positive Economics, University of Chicago Press. Ghosh, A., Gulde, A., Ostry J. and Wolf H., 1995, Does the Nominal Exchange Rate Matter?, IMF Working Paper 95/121. Hadass, Y., Williamson, J., 2001, Terms of Trade and Economic Performance 1870-1940: Prebisch and Singer Revisited, NBER WP 8188. IMF, 1997, Exchange Rate Arrangements and Economic Performance in Developing Countries, Ch. 4 of World Economic Outlook, October. Kenen, P., 2002, Currencies, Crises, and Crashes, Mimeo. Levy Yeyati, E., Sturzenegger, F., 2002, Classifying Exchange Rate Regimes: Deeds vs. Words, Mimeo. Levy Yeyati, E., Sturzenegger, F., 2003, To Float or to Fix: Evidence on the Impact of Exchange Rate Regimes on Growth, American Economic Review, forthcoming. Lutz, M., Singer, H.W., 1994, Trend and Volatility in the Terms of Trade: Consequences for Growth, The Economics of Primary Commodities: Models, Analysis and Policy, W. Morgan and D. Sapsford, pp. 91-121. Meade, J., 1951, The Theory of International Economic Policy, 2 volumes. Mendoza, E., 1997, Terms-of-Trade Uncertainty and Economic Growth, Journal of Development Economics. Obstfeld, M., Rogoff, K., 1995, The Mirage of Fixed Exchange Rates, Journal of Economics Perspectives, Fall. Prebisch, R., 1950, The Economic Development of Latin American and Its Principal Problems , New York, reprinted in Economic Bulletin for Latin America, 7 (1962):1-22.

20

Rose, A., 2000, One Money, One Market: Estimating The Effect of Common Currencies on Trade, Economic Policy. Rose, A., Van Wincoop, E., 2001, National Money as a Barrier to International Trade: The Real Case for Currency Union, American Economic Review, Papers and Proceedings, 91:2, 386-390. Sachs, J.D., Warner, A.M.,1995, Economic Reform and the Process of Global Integration, Briookings Papers on Economic Activity 1: pp. 1-118. Singer, H.W., 1950, Gains and Losses from Trade and Investment in Under- Developed Countries, AER. Warner, A. M., 1992, Did the Debt Crisis Cause the Investment Crisis, Quarterly Journal of Economics, 107:1161-86.

21

Table 1. Distribution of Exchange Rate Regimes

Pegged Non-Pegged

Conventional Hard Total Intermediate Flexible Total Total

1356 717 2073 600 662 1262 3335

Source: Levy Yeyati and Sturzenegger (2001).

Table 2. Summary Statistics

Variable Obs. Mean Std. dev. Min Max

∆gdp 100 1.157 1.933 -5.689 5.850

gdpin 100 1.340 1.690 0.066 9.828

invgdp 100 0.217 0.054 0.097 0.440

sec 100 0.416 0.275 0.020 0.910

openness 100 0.355 0.204 0.083 1.168

gov 100 0.203 0.192 0.024 1.148

civil unrest 1733 3.494 1.745 1 7

∆tt 1733 4.226 16.185 -88.846 114.195

22

Table 3. Cross-Section First-Stage Growth Regressions

(i) (ii) (iii) gdpin -0.550*** -0.467** -0.494** (0.135) (0.204) (0.197) invgdp 6.276 5.171 5.052 (4.288) (4.259) (4.427) sec 2.969*** 2.591** 2.594** (0.908) (1.138) (1.105) openness 0.479 1.351 1.531 (0.970) (1.048) (1.098) gov -2.030 -2.520 -2.234 (1.460) (1.550) (1.514) latam -0.971** -0.828* -0.858* (0.467) (0.461) (0.471) safrica -1.480*** -1.191** -1.280** (0.547) (0.566) (0.554) trans -0.865** -0.557 -0.621 (0.412) (0.615) (0.593) pegged_cross -0.854*** (0.318) regime_cross -0.656** (0.278) constant 0.157 0.588 0.923 (0.950) (0.912) (1.001) Obs. 100 96 96 R2 0.45 0.45 0.44 Note: ***, **, and * represent 99, 95 and 90% significance. Heteroskedasticity-consistent standard errors in parentheses.

23

Table 4.a. Growth Dynamics Regressions (FGLS) Full Sample

(i) (ii) (iii) (iv) (v) (vi) [ g*

j – g t-1 j ] 0.793*** 0.751*** 0.793*** 0.803*** 0.791*** 0.748*** (0.020) (0.023) (0.02) (0.023) (0.02) (0.023) ∆tt 0.043*** 0.051*** 0.058*** 0.060*** 0.037*** 0.045*** (0.008) (0.008) (0.007) (0.007) (0.011) (0.011) ∆tt_1 0.026*** 0.036*** 0.021* (0.009) (0.007) (0.011) ∆tt*pegged 0.037*** 0.037*** (0.011) (0.011) ∆tt*pegged_1 0.018* (0.011) ∆tt*hard 0.03 0.043** (0.018) (0.018) ∆tt*hard_1 0.014 (0.018) ∆tt*regime 0.019*** 0.020*** (0.007) (0.007) ∆tt*regime_1 0.011* (0.007) civil unrest -0.029 -0.018 -0.028 -0.02 -0.034 -0.026 (0.040) (0.040) (0.041) (0.040) (0.040) (0.040) constant -0.182 -1.158** -0.216 -1.165** -0.142 -1.140** (0.488) (0.459) (0.495) (0.472) (0.490) (0.459) Obs. 1733 1650 1733 1723 1733 1650 ∆tt + ∆tt_1 0.077*** 0.096*** 0.066*** [47.90] [111.62] [22.93] Pegged a 0.055*** [15.23] Hard b 0.057* [5.58] Regime c 0.031*** [13.35]

Note: ***, **, and * represent 99, 95 and 90% significance. Heteroskedasticity-consistent standard errors in parentheses. χ2 in brackets. All regressions include year dummies. a Refers to: ∆tt*pegged + ∆tt*pegged_1. b Refers to: ∆tt*hard + ∆tt*hard_1. c Refers to: ∆tt*regime+∆tt*regime_1.

24

Table 4.b. Growth Dynamics Regressions (FGLS) Split Sample by Regime

(i)

Flexible (ii)

Intermediate (iii) Peg

(iv) Hard Peg

[ g* j – g t-1 j ] 0.882*** 0.938*** 0.767*** 0.909***

(0.035) (0.035) (0.030) (0.066) ∆tt 0.037*** 0.048*** 0.083*** 0.125*** (0.009) (0.010) (0.008) (0.020) ∆tt_1 0.020* 0.023** 0.046*** 0.043** (0.009) (0.010) (0.008) (0.020) civil unrest 0.084 -0.116 -0.02 -0.108 (0.060) (0.082) (0.052) (0.251) constant -1.236** 2.109 0.001 1.494 (0.603) (1.567) (0.386) (2.126) Obs. 462 416 845 225 ∆tt + ∆tt_1 0.057*** 0.071*** 0.129*** 0.168*** [17.76] [25.06] [134.28] [39.68] Note: ***, **, and * represent 99, 95 and 90% significance. Heteroskedasticity-consistent standard errors in italics. χ2 in brackets. All regressions include year dummies.

Table 4.c. Differential Response by Regime: χ2 Tests

Flexible Intermediate Hard

Pegged 0.068*** a

[9.17] 1.294*** b

[6.77] -0.064*c [3.49]

Obs. 1307 1261 845 Notes: ***, **, and * represent 99, 95 and 90% significance. χ2 in brackets. All regressions include year dummies. a Refers to: ∆tt*pegged + ∆tt_1*pegged – (∆tt*flexible + ∆tt_1*flexible). b Refers to: ∆tt*pegged + ∆tt_1*pegged – (∆tt*intermediate + ∆tt_1*intermediate). c Refers to: ∆tt*(pegged-hard) + ∆tt_1*(pegged-hard) – (∆tt*hard + ∆tt_1*hard).

25

Table 5. Growth Dynamics Regressions (FGLS) Emerging and Industrial Countries

(i) Emerging

(ii) Industrial

(v) Emerging

(vi) Industrial

[ g* j – g t-1 j ] 0.789*** 0.639*** 0.787*** 0.638***

(0.027) (0.045) (0.027) (0.045) ∆tt 0.049*** 0.057*** 0.036*** 0.071*** (0.010) (0.019) (0.013) (0.020) ∆tt_1 0.025*** 0.012 0.024* -0.006 (0.010) (0.019) (0.013) (0.020) ∆tt*pegged 0.046*** -0.013 (0.012) (0.027) ∆tt*pegged_1 0.016 0.068** (0.012) (0.027) ∆tt*regime 0.028*** -0.018 (0.008) (0.014) ∆tt*regime_1 0.008 0.049*** (0.008) (0.014) civil unrest -0.122* -0.338** -0.125* -0.348** (0.064) (0.144) (0.064) (0.142) constant -0.141 -2.171*** 0.533 0.352 (0.512) (0.450) (0.559) (0.437) Obs. 1281 369 1281 369 ∆tt + ∆tt_1 0.074*** 0.069*** 0.060*** 0.065*** [33.24] [10.27] [12.09] [12.59] Pegged a 0.062*** 0.055** [15.64] [7.31] Regime b 0.036*** 0.031*** [13.62] [13.09] Note: ***, **, and * represent 99, 95 and 90% significance. Heteroskedasticity-consistent standard errors in parentheses. χ2 in brackets. All regressions include year dummies. a Refers to: ∆tt*pegged + ∆tt*pegged_1. La b Refers to: ∆tt*regime+∆tt*regime_1.

26

Table 6. Asymmetry (FGLS) Full Sample and Emerging Countries

(i) (ii) (iii) (iv) (v) (vi) (vii) All Emerging Flexible Intermediate Peg Hard Flexible Intermediate Peg [ g*

j – g t-1 j ] 0.883*** 0.916*** 0.767*** 0.907*** 0.926*** 0.974*** 0.803*** (0.035) (0.037) (0.030) (0.067) (0.039) (0.043) (0.033) ∆tt*pos 0.022 0.029** 0.066*** 0.070* 0.032* 0.031* 0.062*** (0.014) (0.015) (0.013) (0.037) (0.016) (0.018) (0.015) ∆tt*pos_1 0.041*** 0.004 0.040*** 0.046 0.027* -0.006 0.029** (0.014) (0.015) (0.012) (0.033) (0.016) (0.018) (0.013) ∆tt_neg 0.060** 0.067*** 0.105*** 0.173*** 0.065** 0.079*** 0.119*** (0.024) (0.022) (0.015) (0.032) (0.027) (0.023) (0.016) ∆tt*neg_1 -0.007 0.047** 0.053*** 0.031 0.013 0.059*** 0.055*** (0.017) (0.020) (0.016) (0.035) (0.018) (0.022) (0.016) civil unrest 0.095 -0.024 0.020 -0.061 0.084 -0.092 -0.124 (0.065) (0.087) (0.059) (0.249) (0.119) (0.128) (0.084) constant -1.352** 2.103 0.202 1.777 -1.791 1.579 1.267 (0.577) (1.593) (0.398) (2.151) (1.156) (2.001) (0.826) Obs. 462 416 845 225 301 326 714 pos b 0.063*** 0.033 0.106*** 0.116** 0.059*** 0.025 0.091*** [12.00] [2.49] [40.69] [5.82] [7.99] [0.92] [25.30] neg c 0.053* 0.114*** 0.158*** 0.204*** 0.078** 0.138*** 0.174*** [3.45] [16.98] [58.13] [19.80] [5.99] [20.42] [64.79] neg – pos d -0.010 0.081** 0.052* 0.088 0.019 0.113*** 0.083*** [0.08] [4.86] [3.26] [1.70] [0.24] [6.82] [7.16] Notes: ***, **, and * represent 99, 95 and 90% significance. Heteroskedasticity-consistent standard errors in parentheses. χ2 in brackets. All regressions include year dummies. a There are no hard pegs among industrial countries. b Refers to: ∆tt*pos + ∆tt*pos_1. c Refers to: ∆tt*neg + ∆tt*neg_1. d Refers to: ∆tt*pos+∆tt*pos_1 - ∆tt*neg - ∆tt*neg_1.

27

Table 7. Asymmetry (FGLS) Differential Response by Regime and Type of Shock: χ2 Tests

Pegged – Flex Pegged – Nonpegged Flex – Nonflex

Positive shock 0.032 [1.18]

0.060*** [5.81]

0.044 [2.02]

Negative shock 0.103*** [5.70]

0.080** [5.34]

0.086* [3.38]

Obs. 1307 1723 1723

Notes: ***, **, and * represent 99, 95 and 90% significance. χ2 in brackets. All regressions include year dummies.

28

Appendix A. Description of the Data (a) Variables and Sources Variable Definitions and sources g Rate of growth of real per capita GDP (Source: World Economic Outlook [WEO])

∆tt Change in terms of trade - exports as a capacity to import (constant LCU) (Source: World Development Indicators [WDI]; variable NY.EXP.CAPM.KN)

civil unrest Index of civil liberties (measured on a 1 to 7 scale, with one corresponding to highest degree of freedom) (Source: Freedom in the World - Annual survey of freedom country ratings)

gdpin Initial per capita GDP (average over 1970-1973) (Source: WEO)

gov Growth of government consumption (Source: IMF’s International Financial Statistics [IMF])

invgdp Investment to GDP ratio (Source: IMF) openness Openness, (ratio of [export + import]/2 to GDP) (Source: IMF). sec Total gross enrollment ratio for secondary education (Source: Barro, 1991) latam Dummy variable for Latin American countries safrica Dummy variable for Sub-Saharan African countries trans Dummy variable for Transition economies

29

(b) List of Countries (183-country sample; industrial countries in bold) Australia Burkina Faso Jamaica Philippines Austria Burundi Jordan Poland Belgium Cambodia Kazakhstan Qatar Canada Cameroon Kenya Romania Denmark Cape Verde Kiribati Russia Finland Central African Rep. Korea Rwanda France Colombia Kuwait Samoa Germany Comoros Kyrgyz Republic Sao Tome & Principe Greece Congo, Dem. Rep. Of Lao People's Dem.Rep Saudi Arabia Iceland Congo, Republic Of Latvia Senegal Ireland Costa Rica Lebanon Seychelles Italy Cote D Ivoire Lesotho Sierra Leone Japan Croatia Liberia Singapore Netherlands Cyprus Libya Slovak Republic New Zealand Czech Republic Lithuania Slovenia Norway Chad Luxembourg Solomon Islands Portugal Chile Macedonia, Fyr Somalia San Marino China,P.R.: Mainland Madagascar South Africa Spain China,P.R.:Hong Kong Malawi Sri Lanka Sweden Djibouti Malaysia St. Kitts And Nevis Switzerland Dominica Maldives St. Lucia United Kingdom Dominican Republic Mali St. Vincent & Grens. United States Ecuador Malta Sudan Afghanistan, I.S. Of Egypt Marshall Islands Suriname Albania El Salvador Mauritania Swaziland Algeria Equatorial Guinea Mauritius Syrian Arab Republic Angola Estonia Mexico Tajikistan Antigua And Barbuda Ethiopia Micronesia, Fed.Sts. Tanzania Argentina Fiji Moldova Thailand Armenia Gabon Mongolia Togo Aruba Gambia, The Morocco Tonga Azerbaijan Georgia Mozambique Trinidad And Tobago Bahamas, The Ghana Myanmar Tunisia Bahrain Grenada Namibia Turkey Bangladesh Guatemala Nepal Turkmenistan Barbados Guinea Netherlands Antilles Uganda Belarus Guinea-Bissau Nicaragua Ukraine Belize Guyana Niger United Arab Emirates Benin Haiti Nigeria Uruguay Bhutan Honduras Oman Vanuatu Bolivia Hungary Pakistan Venezuela, Rep. Bol. Bosnia And Herzegovina India Palau Vietnam Botswana Indonesia Panama Yemen, Republic Of Brazil Iran, I.R. Of Papua New Guinea Zambia Brunei Darussalam Iraq Paraguay Zimbabwe Bulgaria Israel Peru

30

Appendix B. De Facto Exchange Rate Regime Classification 18

The de facto classification of exchange rate regimes used in this paper employ

cluster analysis techniques to group countries according to the behavior of three variables

related to exchange rate policy: (i) Exchange rate volatility (σe), measured as the average

of the absolute monthly percentage changes in the nominal exchange rate relative to the

relevant anchor currency (or basket of currencies, whenever the currency weights are

disclosed) over the year; (ii) Volatility of exchange rate changes (σ∆e), measured as the

standard deviation of the monthly percentage changes in the exchange rate; and (iii)

Volatility of reserves (σr), measured as the average of the absolute monthly change in

dollar denominated international reserves relative to the dollar value of the monetary base

in the previous month.

These variables are computed on an annual basis, so that each country-year

observation represents a point in the (σe, σ∆e, σr) space. In this space, floats are

associated with little intervention in the exchange rate market (low volatility of reserves)

together with high volatility of exchange rates. Observations with little or no exchange

rate volatility and substantial reserves volatility correspond to the group of fixes. Finally,

intermediate regimes are associated with moderate to high volatility across all variables,

reflecting exchange rate movements in spite of active intervention. Observations are

grouped by proximity using cluster analysis according to the characteristics previously

identified.19

18 Based on Levy Yeyati and Sturzenegger (2002). 19 Those that do not display significant variability in either dimension are judged “inconclusives,” and left unclassified.